Frontier AI labs still won’t say how they’d contain a rogue model

The labs assessed were Anthropic, Google, OpenAI, Meta, and xAI. Guidelight found that the labs still will not say how they would contain a rogue model. The assessment relied on information the labs have made public rather than internal documents. The article refers to the assessed companies as frontier AI labs.

The evaluation graded four specific areas. These are logging and monitoring, whether systems are halted after flagged misbehavior, third-party audits, and exact containment plans. Regarding logging and monitoring, the assessment looked at how the labs record and oversee model behavior. On the question of halting, it examined whether the labs commit to stopping systems when a problem is flagged. For third-party audits, it considered whether outside reviewers have access. The assessment found that the labs have not disclosed exact containment plans in their public materials. The article does not provide individual scores for each lab, nor does it rank the labs against one another.

The article was published on August 22, 2026, and is titled "Frontier AI labs still won't say how they'd contain a rogue model." It notes that the labs have not given specific public answers on how they would handle a rogue model. The assessment is described as covering current publicly available plans. No responses from the labs to the assessment are included in the excerpt. The article does not specify what consequences, if any, might follow from the assessment's findings.